gyan/doc/source/developer-guide.md

Developer Guide

Openstack Gyan is the Machine Learning Infra as a service project. This document provides the following

• How to setup openstack with Gyan using devstack
• Describes the workflow in the Gyan
• How to run the example application provided in the Gyan examples directory

How to setup Gyan

• Clone the devstack master branch

  git clone https://github.com/openstack-dev/devstack
  

• Copy the contents below to local.conf in devstack

  [[local|localrc]]
  HOST_IP=192.168.1.188
  SERVICE_HOST=192.168.1.188
  DATABASE_PASSWORD=password
  RABBIT_PASSWORD=password
  SERVICE_TOKEN=password
  SERVICE_PASSWORD=password
  ADMIN_PASSWORD=password
  enable_service rabbit mysql key
  enable_plugin gyan https://git.openstack.org/openstack/gyan
  enable_plugin heat https://git.openstack.org/openstack/heat
  LIBS_FROM_GIT="python-gyanclient"
  ENABLED_SERVICES+=gyan-api
  

• Run stack.sh

  ./stack.sh
  

Workflow of Gyan:

• Create hints.yaml to specify the size of the compute node. For example

    python_version: 2.7
    cpu: 2
    memory: 1024 Mb
    disk: 20 Gb
    driver: TensorflowDriver
    additional_details: {}
  

• After that create the flavor using hints template

  gyan flavor-create --hints-path flavor.yaml tensorflow
  

• Compress your ML model that is already trained and create a Gyan model

  gyan create-model --trained-model model.zip --type Tensorflow MNIST
  

• Deploy model that was created in the previous step.

  gyan deploy-model <model-id>
  

• The above command will launch the compute node based on the flavor details we gave in the first step. Once the compute node is launched, the gyan-compute will be installed and connected to gyan-server.
• We should get new host in gyan host-list.
• In the last step you should see the deployed url of the model. We can find out using gyan model-list

How to use run sample example KnowThyNumber provided with Gyan:

• Make sure you have installed golang in your system and set the GOROOT and GOPATH properly.
• Copy the KnowThyNumber example from gyan/examples to GOPATH/src/github.com/<user-id>/KnowThyNumber
• Now run go run server.go. This will start local server on 9000 port.
• We got deployed_url in the previous section. Get the openstack token using openstack token issue.
• Open the browser and navigate to http://localhost:9000. Provide the deployed_url and token in the app. Now you can draw any number in the canvas and use your model to predict it.